Can you clarify what "sample materials" (left) and "sample lights" (right) mean? Do they correspond to certain importance densities, as in the previous slide? Is this still applying bidirectional path tracing, as in slide 32? (http://15462.courses.cs.cmu.edu/fall2019/lecture/variance/slide_032)

motoole2

@Kalecgos "Sample materials" and "sample lights" are two different sampling strategies to generate a light path from a source, to a point in the scene, and then to the camera. In both cases, we start by shooting a ray from the camera to a point in the scene. The objective is to connect this point to the light source.

In "sample materials", we randomly sample a new direction based off of the material's BRDF. This works well for specular BRDFs where there's only one ray to pick from, but requires multiple samples for glossy/diffuse BRDFs before most rays miss the light source. For "sample lights", we randomly pick a point on the light source and draw a ray between the point in the scene and this point on the light source. This guarantees that our light path reaches our light source, but ignores the BRDF of the scene point. For specular materials and large light sources, we get noisy results because the line segment joining the scene point and the light source has a BRDF value of 0.

Can you clarify what "sample materials" (left) and "sample lights" (right) mean? Do they correspond to certain importance densities, as in the previous slide? Is this still applying bidirectional path tracing, as in slide 32? (http://15462.courses.cs.cmu.edu/fall2019/lecture/variance/slide_032)

@Kalecgos "Sample materials" and "sample lights" are two different sampling strategies to generate a light path from a source, to a point in the scene, and then to the camera. In both cases, we start by shooting a ray from the camera to a point in the scene. The objective is to connect this point to the light source.

In "sample materials", we randomly sample a new direction based off of the material's BRDF. This works well for specular BRDFs where there's only one ray to pick from, but requires multiple samples for glossy/diffuse BRDFs before most rays miss the light source. For "sample lights", we randomly pick a point on the light source and draw a ray between the point in the scene and this point on the light source. This guarantees that our light path reaches our light source, but ignores the BRDF of the scene point. For specular materials and large light sources, we get noisy results because the line segment joining the scene point and the light source has a BRDF value of 0.